All pull requests

Pull requests

• mamba-ssm: test ROCm parallel arch build

  chore
  #1331·danieldk opened Nov 13, 2026Member·

  ·
  3

• ci: enable MPS runners to test metal kernels

  ci
  #1330·sayakpaul opened Nov 13, 2026Member·

  ·
  3

• fix: activation gelu_fast rounding and empty launches, mamba-ssm generate output

  fix
  #1329·drbh opened Nov 12, 2026Collaborator·

  ·
  3

• deep-gemm: fused per-token FP8 cast in Triton (fused_per_token_cast_to_fp8), version 3

  chore
  #1326·IlyasMoutawwakil opened Nov 12, 2026Member·

  ·
  8

• flydsl: SDK experiment

  chore
  #1323·danieldk opened Nov 11, 2026Member·

  ·
  5

• feat: generate einops at build time instead of committing src

  feature
  #1322·drbh opened Nov 11, 2026Collaborator·

  ·
  9

• cv-utils: build for rocm

  chore
  #1320·drbh opened Nov 10, 2026Collaborator·

  ·
  4

• causal-conv1d: build for rocm

  chore
  #1319·drbh opened Nov 10, 2026Collaborator·

  ·
  5

• uccl: as an sdk variant

  chore
  #1316·danieldk opened Nov 10, 2026Member·

  ·
  3

• flash-attn2: [XPU] split-KV decode kernel for seqlen_q == 1

  chore
  #1309·jiqing-feng opened Nov 9, 2026Contributor·

  ·
  1

• feat(tokamax): add narrow tokamax 0.0.17 TPU kernel

  feature
  #1307·tengomucho opened Nov 6, 2026Member·

  ·
  7

• Cv utils trim rows

  chore
  #1288·guarin opened Oct 30, 2026Member·6 tasks·

  ·
  1

• uccl: experiment with uccl packaged as a kernel

  chore
  #1287·danieldk opened Oct 30, 2026Member·

  ·
  6

• feat: generate flash-attn3 from a kernel-port recipe

  feature
  #1283·drbh opened Oct 28, 2026Collaborator·

  ·
  5

• build(deps): bump the actions group across 1 directory with 8 updates

  dependenciesdepsgithub_actions
  #1282·dependabot[bot] opened Oct 28, 2026Bot·

  ·

• finegrained-kernels: add the missing quantization formats and generalize kernel optimizations

  choreneeds-rebase
  #1240·IlyasMoutawwakil opened Oct 16, 2026Member·

  ·
  1

• deep-ep: add DeepEP 2.5.0 (expert-parallel dispatch/combine)

  chore
  #1231·IlyasMoutawwakil opened Oct 15, 2026Member·3 of 6 tasks·

  ·
  35

• ragged-dot-tpu: add tokamax MoE experts kernel for TPU

  chore
  #1208·tengomucho opened Oct 8, 2026Member·3 of 6 tasks·

  ·
  12

• gemma-quant-kernels: add fused INT2/INT4/INT8 GEMV, GEMM, and MoE grouped GEMM

  chore
  #1192·pculliton opened Sep 29, 2026·3 of 6 tasks·

  ·
  11

• fix(ci): harden GitHub Actions workflows (#1169)

  fix
  #1170·hf-security-analysis[bot] opened Sep 22, 2026Bot·

  ·

• finegrained-fp8: bound the grouped kernels to N and K

  chore
  #1154·malaiwah opened Sep 16, 2026·

  ·
  1

  5

• rotary: support partial RoPE in apply_rotary_transformers

  chore
  #1143·piyushumate opened Sep 9, 2026·

  ·
  29

• vllm-quantization: import in kernels-community again

  cudafeaturenew-kernelrocm
  #991·danieldk opened Jun 23, 2026Member·

  ·

• tokenspeed-attention: add TokenSpeed MHA/MLA Triton kernels

  featurenew-kerneltritonupstream-sync
  #990·adarshxs opened Jun 23, 2026Contributor·

  ·
  2

• make ScatterMoe more easily patchable for HF MoE modeling

  cudafeaturenew-layer
  #82·winglian opened Dec 5, 2025·

  ·
  12